【作者】 杨东梅；

【导师】 印桂生；

【作者基本信息】 哈尔滨工程大学 ， 计算机应用技术， 2010， 博士

【摘要】 虚拟装配技术为制造业带来了全新的设计理念,从本质上将传统制造从设计到生产的不断修改、多次试制的过程中解脱出来。装配工艺规划是虚拟装配最核心的部分,本文侧重基于智能计算的虚拟装配工艺规划进行深入研究,主要包括两部分：装配序列规划和装配路径规划。论文的主要工作如下：针对虚拟装配过程的复杂性,提出面向装配规划的产品层次信息模型,将零件模型信息依次存储在零件属性层、面片显示层、装配关系层及过程信息层,各个层次间通过零件索引号进行数据间的约束与映射,从而实现模型信息层级间的相互关联。以装配语义形式描述装配过程中零件间的配合约束,建立干涉矩阵和线性自由度矩阵描述零件间的联接关系；同时,为装配模型引入过程信息描述,以得到装配序列、路径等动态描述信息,并借助过程信息,对零件模型的装配关系进行精确的评价。层次信息模型利于装配系统依据不同的装配任务对各个层级的模型信息进行读取,从而提高装配效率。针对虚拟装配序列求解过程中出现的“组合爆炸”现象,在建立拆卸干涉矩阵的基础上,提出改进蚁群算法进行装配序列求解,只有在一次迭代循环中找到最优拆卸序列的蚂蚁在相应的路径上增加全局信息素；选取蚂蚁的个数等于初始可行拆卸操作的数目。对于具有较强约束的零部件,强约束条件减少了算法初始可行的拆卸操作,局限了选择问题的解空间,算法效率明显提高。针对结构较复杂的装配体,结合遗传算法和蚁群算法求解装配序列的特点,提出遗传蚁群混合算法求解最优装配序列。算法的主要思想：每当蚂蚁完成一次漫游后,将蚂蚁构建的可行序列加入遗传算法的初始种群,遗传算法对该可行装配序列进行全局优化,并依据优化后生成解的质量在对应的蚂蚁爬行路径上释放相应浓度的信息素,如此循环交叉调用遗传算法和蚁群算法,使遗传蚁群算法求精能力显著增强。提出遗传算法与栅格法结合进行装配路径规划,采用栅格表示装配体的初始位置及装配空间环境地图,栅格路径的序号而不是传统的二进制作为种群个体的编码,适应度函数转换为寻找装配体的最优装配路径,保证了虚拟装配过程中装配体的路径为最优安装路径,提高算法的搜索效率,同时有效避免了传统搜索算法的局部极小值问题。构建一个舱段虚拟装配原型系统,提出该系统的构建思想和体系结构,并采用内存调度策略、多线程的运动控制完善装配系统性能。采用基于相对位置的碰撞干涉剔除,由装配体当前位置、移动方向、相对位置之间的关系来约束装配体的位置变换,从而完成装配体安装；提出基于径向基神经网络的场景调度策略,将虚拟化身当前视点状态作为网络的输入样本,利用径向基神经网络预测当前视点的后续状态。获得当前视点的后续状态后,结合视锥体进行取景,即可进行场景调度。漫游过程中化身以“沉浸”方式身临其境体验装配工艺规划方案,使装配过程更真实地回溯。更多还原

【Abstract】 The virtual assembly has brought the newly design for the industry of manufacturing, and the technology relief the traditional process from design, production and constantly trial process. Assembly process planning is the most crucial part of the virtual assembly. This paper focuses on the deeply study in the area of virtual assembly process planning based on intelligence computation. The main thesis is as follows:For virtual assembly process complexity, proposed the product-hierarchical information model for the assembly planning. The informations of part model are stored in sequence in the parts attribute layer, the surface layer of film shows, the assembly relation layer, the information process layer, these layers implementation the linkages between information-model level by the data constraints and mapping by part index number. The constraint among parts assembling process is described as the assembly semantics, build interference matrix and the linear degrees of freedom matrix to describe the relations between parts. Using information describing the process in the assembly model for acquire the assembly sequence, path and other dynamic descriptions. Evaluate the relationship between the part model with the process information. Hierarchy information model convenience the assembly system according to different tasks to operate at different information model.To solving the combinatorial explosion phenomenon during the solving process of virtual assembly sequence, this paper proposed a method for obtain the assembly sequence using ant colony algorithm based on the establishment of the demolition interference matrix. Only the ant which has found the optimal disassembly sequence in once iteration cycle could release global pheromone in the corresponding path; the number of ants is equal to the number of initial feasible demolition operations. Constraints between parts reduce the demolition times of the initial algorithm operation, limit the selection of the solution space, improve the efficiency of the algorithm.Combining the characteristics of genetic algorithm and ant colony algorithm for assembly sequence, this paper proposed Hybrid Genetic-Ant Colony Algorithm for optimal assembly sequence. The main idea of the algorithm followed:When the ants after a tour, the feasible seqence which is constructed by ants as a part of initial population of genetic algorithm. Then the genetic algorithm global optimized the feasible assembly sequence constructed by ants, according to the quality of solution release corresponding concentration of pheromone in the corresponding ant crawling path. Cycle crossover called the genetic algorithm and ant colony algorithm to increase solving ability of the ant colony algorithm.This paper proposed a method with genetic algorithm and grid for assembly path planning. The method used grid to describe the initial location and the space environment map of assembly object. Adopt the serial number of the grid path as the population of individual coding, but not the traditional binary code. The fitness function convertion is used for find the optimal assembly path, it ensured the path of virtual assembly process is the optimal one. It’s improve the search efficiency, and avoid the local minimum problems of the traditional search algorithms.Built a system of thruster module virtual assembly, proposed construction of the system and architecture adopt memory scheduling strategy, multi-threaded motion control to improve the performance of assembly systems. Adopt the collision interference removed based on the relative position, then complete the installation of assembly by restrict the assembly location changes. Proposed a scene scheduling strategy based on RBF neural network, use the current the state point of virtual avator as the input sample, then forecast the current status of the follow-up point of view through RBF neural network. Once the current status of the follow-up point of view is obtained, scheduling scenarios can be carried out through the combination of visual cone for shooting. The avator deeply observe the assembly process plan in the wandering process, this method made the assembly process could be truly look back upon.更多还原